Harvestry is here.
We built Harvestry because rewatching a lecture to find one slide is a terrible use of an hour. Today, version 1.0 is available — and it turns any lecture video into a polished, interactive study document in minutes, entirely on your Mac.
The idea is simple: import a video, press Begin Processing, and Harvestry transcribes the audio with WhisperKit (running on the Apple Neural Engine, never touching the cloud), captures the key screenshots using scene-change detection, and weaves them together into a self-contained HTML document with live audio sync, per-word highlighting, and dark mode built in.
What took a semester of rewatching now takes the length of the video itself — usually less, because Harvestry runs transcription and screenshot capture in parallel using different hardware. While WhisperKit is working through the audio on the Neural Engine, AVFoundation is scanning the video track for scene changes on the GPU. By the time both finish, your document is ready.
What's in v1.0
- On-device transcription.Five Whisper model sizes from Tiny to Large Turbo. No audio ever leaves your Mac.
- Intelligent screenshot capture.Scene-change detection fires on genuine content transitions, not on presenter movement. Grid-based blur analysis using Variance of Laplacian rejects any frame where the slide content is soft — even if a sharp presenter face is in the corner.
- Manual curation tools.A full scrubber timeline, Seek to Clear (steps to the nearest sharp frame in either direction), Add to Transcript (captures any frame at full resolution), inline notes, and title image selection.
- Beautiful HTML export.Warm editorial design with Instrument Sans headings and Newsreader body text. Live audio sync with per-word highlight. Text annotations in four colours, persisted to localStorage. Dark mode. Font size slider. Works offline, forever, with no dependencies.
- URL import.Paste a YouTube, Vimeo, or any of 1,000+ platform URLs and Harvestry downloads and processes it directly. Powered by yt-dlp, managed inside the app.
- Five languages.English, German, Italian, Spanish, and French — fully localised interface and export output.
- Privacy by design.No account, no telemetry, no crash reporting, no usage analytics. The app processes your content locally and minds its own business.
Coming next
- LLM consolidation (Standard).Send your transcript to the Claude API to produce a refined, consolidated version of your notes alongside the verbatim record.
- LLM consolidation via Ollama (Pro).Run the same consolidation entirely on-device using any locally installed Ollama model — no data leaves your Mac.
- Configurable consolidation prompt.Tell the model what to emphasise, what to skip, and how to structure the output.